A Practical Guide to Measuring Incremental Impact in Food Marketing
The campaign is over.
Sales are up 18% versus the previous four weeks.
Paid media platforms report strong performance.
Social content generated millions of impressions.
Retail media dashboards attribute a substantial amount of revenue to advertising.
ROAS looks healthy.
The obvious conclusion is:
“The campaign worked. Sales increased by 18%.”
But one question has not yet been answered:
What would sales have been if the campaign had never happened?
During the same period:
distribution may have expanded;
the product may have been discounted;
seasonality may have increased category demand;
a major competitor may have gone out of stock;
shelf placement may have improved;
a holiday may have changed consumption;
retail availability may have increased;
or the entire category may simply have grown.
In those circumstances, sales increased while the campaign was running.
That does not automatically mean they increased because of the campaign.
This is the distinction between reporting performance and measuring actual effectiveness.
The Hard Part of Marketing Measurement Is Not Sales. It Is Causality.
Almost every food brand can calculate how much it sold during a campaign.
The harder question is:
How much of that sales volume would have happened anyway — and how much was genuinely created by the campaign?
The second number is what marketers usually mean by incrementality.
IAB and IAB Europe’s 2025 guidance on incremental measurement in commerce media centres the problem around three ideas:
a credible estimate of what would have happened without advertising;
control of measurement bias;
and separation of real effects from statistical noise.
Behind the technical language is a very simple business question:
What happened because we intervened?
That question should sit near the centre of campaign measurement.
Four Numbers That Should Never Be Treated as Interchangeable
Before discussing methodology, it helps to separate four different metrics.
1- Total Sales
Everything sold during the campaign period.
This number is commercially important.
It does not tell us how much advertising caused.
2- Attributed Sales
Sales assigned to advertising according to a platform, retailer or attribution rule.
For example, a customer may:
see an ad;
click an ad;
or purchase within a defined attribution window.
The platform then assigns that sale to the campaign.
This is useful for understanding customer paths and campaign activity.
But it does not answer:
Would this customer have purchased anyway?
3- Incremental Sales
Sales estimated to have occurred because of the campaign.
Circana describes sales lift in similar terms: the objective is to separate genuinely advertising-driven purchases from purchases that would likely have happened regardless.
This is the number marketers often intend to discuss when they say:
“The campaign generated sales.”
But many campaign reports actually show attributed sales instead.
That distinction matters.
4- Incremental Profit
Even incremental sales are not necessarily the end of the analysis.
A campaign may create additional revenue but still destroy short-term economic value after accounting for:
- media;
- discounts;
- retailer fees;
- sampling;
- agency costs;
- creative production;
- and other campaign expenses.
So the commercially relevant question may ultimately become:
How much incremental economic value did we create?
Not simply:
How much revenue appeared in the campaign report?
A Simple Example Shows Why This Matters
Suppose a food brand spends $200,000 on advertising.
The media platform reports:
$1.2 million in attributed sales.
On the surface:
ROAS = 6x
That sounds excellent.
But a controlled experiment or credible model estimates that around $900,000 of those sales would probably have occurred even without advertising.
The truly incremental sales are therefore closer to:
$300,000
Now incremental ROAS is closer to:
1.5x
rather than 6x.
Suppose the product’s gross margin is 35%.
That $300,000 in incremental revenue generates roughly $105,000 in gross margin before other campaign costs.
The interpretation changes dramatically.
Of course, the analysis still may not be complete.
The campaign could have recruited new households that later repurchase.
It could have created longer-term brand effects.
But the example illustrates a fundamental point:
The same campaign can look spectacular or mediocre depending on what is actually being measured.
Why Food Campaign Measurement Is Particularly Difficult
Food sales are influenced by many variables that move at the same time as advertising.
That makes causal measurement harder.
Price
Even modest changes in price can materially affect volume.
Promotion
Discounts, coupons, bundles and multi-buy offers may run at the same time as paid media.
Distribution
If a product expands from 2,000 stores to 3,800 stores during the campaign, higher sales should not automatically be credited to advertising.
NIQ’s 2026 guidance on retail-media incrementality makes this point explicitly: without distribution and availability data, media effects can easily be confused with changes in access.
Seasonality
Ice cream, soft drinks, chocolate, holiday foods, Ramadan products and cooking ingredients all have different demand cycles.
Availability
Advertising can generate demand while stockouts prevent that demand from becoming sales.
Competitive Activity
A competitor may:
change price;
run a major promotion;
launch a new product;
or experience an out-of-stock period.
All can affect the measured outcome.
Shelf Execution
Additional facings, end-cap placement or stronger merchandising can increase sales independently of media.
Stockpiling
A heavy promotion may cause shoppers to buy three packs today instead of one, without increasing long-term consumption.
Cannibalisation
A campaign may increase sales of Product A partly by reducing sales of Product B from the same brand.
If these variables are not recorded, the campaign can receive credit for effects it did not create.
To Estimate Real Impact, We Need a World Without the Campaign
The problem is that this world cannot be directly observed.
A market cannot simultaneously:
receive the campaign and not receive the campaign.
So analysts must estimate the second condition.
In causal inference, this unobserved scenario is called the counterfactual.
In plain language:
What would most likely have happened if we had done nothing?
The conceptual equation is simple:
Observed sales − estimated sales without the campaign = incremental sales
The difficult part is estimating the second number credibly.
That is where measurement design matters.
Before-and-After Analysis Is Useful — but It Does Not Prove Causality
A very common analysis looks like this:
Sales before the campaign: $10 million.
Sales during the campaign: $12 million.
Conclusion:
“The campaign generated 20% growth.”
That conclusion may be wrong.
Between those periods:
- price;
- distribution;
- holidays;
- promotion;
- availability;
- competition;
- and category demand
may all have changed.
Before-and-after analysis is still useful.
It tells us:
what changed.
It does not, by itself, tell us:
what caused the change.
This is one of the most important distinctions in professional campaign analysis:
Descriptive analysis tells us what happened. Causal analysis tries to determine why it happened.
Year-on-Year Comparison Is Not Automatically a Control Group Either
A team may respond:
“Then we will compare with the same period last year.”
That helps with seasonality.
But conditions can still differ:
prices;
distribution;
retailer mix;
competitors;
inflation;
consumer spending;
holiday timing;
product portfolio;
and promotional intensity.
Last year can be a useful reference.
It is not automatically a valid counterfactual.
Randomised Experiments Remain the Strongest Option When They Are Feasible
From a causal-inference perspective, properly designed randomised experiments remain the reference standard.
Google’s Meridian documentation also treats randomised experimentation as the strongest available evidence when it can be implemented correctly.
The logic is straightforward.
One group receives the campaign.
Another sufficiently similar group does not.
If the groups were comparable beforehand and no major additional differences emerge during the test, the difference in outcomes gives much stronger evidence of causal impact.
For food and FMCG brands, this logic can often be applied through:
stores;
households;
regions;
or geographic markets.
Geo Experiments Can Be Particularly Useful in Food and FMCG
Suppose a brand sells across multiple cities or regions.
Markets with similar historical behaviour can sometimes be assigned to:
test markets and control markets.
The campaign runs in the test markets.
The selected intervention is withheld from control markets.
The difference in performance is then evaluated relative to historical trends.
Google has published extensive research on geographic experiments for advertising measurement, including methods for matching comparable geographies and estimating incremental return on ad spend.
But execution requires discipline.
A credible geo experiment should consider:
national-media spillover;
differences in distribution;
promotion intensity;
inventory;
retailer mix;
historical sales trends;
and local category conditions.
A badly constructed control market can produce a confident answer to the wrong question.
A Food-Sector Example: HiPRO
A Circana case study involving HiPRO yoghurt provides a useful example of multi-layer measurement.
The campaign used:
brand-lift research and a geographic sales experiment.
Four regions were held out from the campaign and used as controls for measuring sales impact.
Circana reported a 7.1% incremental sales lift in that specific case.
That number should not be interpreted as:
“Social campaigns increase food sales by 7.1%.”
It was specific to:
that product;
that campaign;
that market;
that audience;
and that experimental design.
The more valuable lesson is methodological.
The campaign asked two different questions:
Did brand perception change?
and
Did sales increase beyond what would likely have happened anyway?
That is a stronger measurement architecture than simply reporting impressions and attributed purchases.
Sales Can Fall While the Campaign Still Creates Positive Incremental Impact
This can sound counterintuitive.
Imagine sales in the campaign market fall by 2%.
At first glance, the campaign appears unsuccessful.
But sales in comparable control markets fall by 8%.
In that case, the campaign may have prevented part of a broader decline.
Without the intervention, performance might have been materially worse.
So:
negative sales growth
and
negative campaign impact
are not necessarily the same thing.
The Reverse Can Also Happen
Suppose sales in the test market grow by 10%.
Excellent.
But control markets grow by 12% over the same period.
That could indicate that:
the whole category was expanding;
seasonality was favourable;
or another market-wide factor was lifting demand.
The campaign market actually underperformed the expected baseline.
So:
The direction of sales movement and the direction of incremental effect can be different.
That is precisely why causal measurement is valuable.
What If a True Experiment Is Not Possible?
Real-world marketing does not always allow clean randomisation.
The campaign may be national.
Television may spill across markets.
Creators may reach everyone.
Retailers may refuse to maintain a control group.
The number of viable regions may be too small.
In these cases, model-based approaches become more important.
IAB’s 2025 incremental-measurement guidance discusses several broad approaches:
experiments;
modelled counterfactuals;
econometric methods;
and hybrid frameworks.
One possible approach is a synthetic control.
Using information such as:
historical sales;
matched markets;
category trends;
pricing;
promotions;
seasonality;
and other relevant variables,
the model estimates what the treated market would probably have done without the campaign.
Research published in the Annals of Applied Statistics on Bayesian structural time-series methods follows this general logic: build a plausible counterfactual from related time series, then compare it with what actually happened after the intervention.
Useful?
Yes.
Magic?
No.
The credibility of the estimate depends on the quality of the model and the information available.
Where Marketing Mix Modelling Fits
Marketing Mix Modelling, or MMM, attempts to estimate how different drivers contribute to a business outcome such as sales.
Potential variables might include:
television;
social media;
search;
retail media;
price;
promotion;
distribution;
seasonality;
macroeconomic conditions;
and competitive activity.
MMM can be valuable for questions such as:
Which channels appear to contribute most over time?
How should budgets be allocated?
What might happen if investment in one channel increases?
How do media and non-media variables interact?
Google’s Meridian framework explicitly approaches MMM as a causal-inference problem, but also emphasises that the credibility of causal conclusions depends on:
model assumptions;
selection of confounders;
data quality;
and, when possible, calibration against experimental evidence.
So MMM should not be treated as a machine that automatically produces the truth.
It is a model.
Its output is only as credible as the structure underneath it.
Experiments and Models Are Complements, Not Competitors
A mature measurement system should not search for one “winning methodology”.
Different tools answer different questions.
Experiments
Can provide strong causal evidence about a specific intervention in a defined period.
Marketing Mix Models
Can provide a broader picture across channels, pricing, promotions and longer time periods.
Brand-Lift Studies
Can examine changes in:
awareness;
ad recall;
brand association;
consideration;
or purchase intent.
Retail Data
Can show actual transactional behaviour.
Repeat-Purchase Analysis
Can help determine whether the buyers generated by the campaign were commercially valuable beyond the first purchase.
The better question is not:
“Which method is correct?”
It is:
“What level of evidence do we need for the decision we are trying to make?”
ROAS Is Useful — but Only If We Know What the Numerator Means
Return on ad spend is one of the most popular campaign metrics.
The formula is simple:
revenue / advertising cost
The complication is that “revenue” can mean very different things.
NIQ’s 2026 retail-media incrementality guidance warns that reported ROAS does not necessarily represent incremental impact and that measurement definitions can vary across retailers and platforms.
IAB and MRC measurement guidance similarly emphasises transparency around:
attribution windows;
measurement definitions;
and comparability across retail-media environments.
So when someone reports:
ROAS = 5
ask:
Five times what?
Total sales?
Attributed sales?
Incremental sales?
Does it include offline purchases?
Which retailers?
What attribution window?
View-through as well as clicks?
Other products from the same brand?
A ROAS number without a measurement definition can look precise while remaining analytically ambiguous.
Last-Click Attribution Can Take Too Much Credit
Imagine this path:
A customer sees a brand video.
Days later, they notice the product in-store.
At home, they search the brand.
They visit a retailer website.
They click a sponsored search result.
Then they buy.
A last-click model may assign almost all credit to the retailer search ad.
But did that ad create demand?
Or did it capture demand created by earlier exposures?
This is the distinction between:
Attribution
Which touchpoint should receive credit for the recorded purchase?
and
Causality
Would the purchase still have happened without this touchpoint?
Those are different questions.
Both can be useful.
Only one directly addresses incrementality.
For Food Brands, “How Much Did We Sell?” Is Still Not Enough
Even after establishing that the campaign created incremental sales, the quality of those sales matters.
Did We Recruit New Buyers?
A campaign may simply encourage loyal customers to purchase a few days earlier.
Or it may recruit households that had never bought the brand before.
Those outcomes are not commercially equivalent.
Circana’s sales-lift frameworks often separate the contribution of new buyers when evaluating incremental growth.
For a penetration-focused campaign, that distinction can be critical.
Did We Grow the Brand — or Cannibalise Our Own Portfolio?
Suppose a campaign generates an additional $500,000 for a new beverage flavour.
But sales of another flavour in the same portfolio fall by $350,000.
The net brand-level effect is much smaller than the new product’s topline result suggests.
This is cannibalisation.
The reverse can also happen.
Advertising Product A may increase sales of Product B.
That is often described as a halo effect.
A rigorous analysis should, where possible, distinguish:
SKU impact;
brand impact;
and category impact.
Did We Create More Consumption — or Pull Future Purchases Forward?
For shelf-stable foods, heavy promotions may cause shoppers to stock up.
A household buys three packs today instead of one.
Campaign-period sales rise sharply.
But the household buys less over the next two months.
Part of the “lift” was simply demand shifted forward in time.
This is why the post-campaign period can be as important as the campaign period itself.
Did the Buyers Come Back?
For product launches, trial matters.
Repeat matters more for durability.
A campaign might generate a large number of first-time buyers who never return.
Another campaign might generate fewer initial buyers but much stronger repurchase.
Those two outcomes have very different implications.
Useful buyer-level questions include:
How many purchasers were new to the brand?
How many had purchased before?
How quickly did they repurchase?
What proportion made a second purchase?
What was their later customer value?
Incremental trial is valuable.
Incremental repeat is usually more revealing.
Brand Effects Should Not Be Reduced to Campaign-Week Sales
This article focuses heavily on incremental sales.
That does not mean every campaign should be judged by immediate revenue.
Some campaigns are designed to:
increase awareness;
change a perception;
enter the consideration set;
build a new association;
or strengthen mental availability.
In these cases, brand-lift studies can examine differences between exposed and control groups on outcomes such as:
awareness;
ad recall;
favourability;
brand association;
and purchase intention.
Kantar and other research providers use this type of design extensively.
But the same caution applies.
Higher purchase intent does not guarantee sales.
Higher awareness does not prove profitability.
The brand metric should correspond to what the campaign was actually designed to change.
Otherwise measurement becomes a search for whichever number improved.
Match the Measurement Window to the Expected Effect
Suppose a campaign is primarily designed to build long-term brand salience.
Measuring only the first week of sales may miss part of its value.
But the opposite error is equally dangerous:
attributing every future sales increase to “long-term brand effects” without evidence.
Kantar’s work on short- and long-term marketing effectiveness argues for evaluating these horizons differently rather than collapsing them into one metric.
The stronger principle is:
The measurement horizon should match the expected effect horizon.
A trial campaign may require a short window.
Repeat may require several purchase cycles.
Brand effects may require a longer observation period.
There is no universally correct window.
Measurement Should Be Designed Before the Campaign
One of the worst times to decide how a campaign will be evaluated is:
after it has ended.
By then:
there may be no control group;
the pre-period data may be incomplete;
distribution changes may not have been logged;
the KPI definition may have changed;
or different teams may begin selecting whichever numbers look most favourable.
A professional measurement plan should define several things before launch.
The Hypothesis
For example:
The campaign will increase trial among households that have not purchased the brand before.
The Primary Outcome
For example:
Incremental sales from new-to-brand buyers.
Supporting Outcomes
Awareness.
Search.
Product-page visits.
Trial.
Repeat.
The Unit of Experimentation
Individual?
Household?
Store?
City?
Region?
The Pre-Period
Long enough to understand normal behaviour and test/control comparability.
The Response Window
Appropriate to the category’s purchase cycle.
A beverage and a large bag of rice should not use identical repeat windows.
Key Confounders
Price.
Promotion.
Distribution.
Availability.
Competition.
Seasonality.
These decisions should exist before the first impression is served.
Do Not Ignore Statistical Power
Suppose the campaign uses:
two small test stores
and
two small control stores.
Daily sales are volatile.
The campaign may genuinely create a 3% lift, but natural variation may be too large to detect it.
A proper test should therefore ask in advance:
What is the smallest effect that matters commercially, and is the experiment large enough to detect it?
This is particularly important in geo experiments, where:
the number of available markets can be limited;
market sizes vary;
and baseline volatility can be high.
Google’s research on geo experimentation deals directly with challenges such as small sample sizes and heterogeneity between markets.
A statistically underpowered experiment may end with the most frustrating but still legitimate conclusion:
We cannot tell.
That is better than claiming certainty where none exists.
Report Uncertainty Instead of Hiding It
Professional measurement should rarely say:
“The campaign effect was exactly 4.37%.”
Experiments and models contain uncertainty.
A stronger report might say:
Our best estimate is approximately 5% incremental lift, with a plausible range around that estimate.
If that interval is wide and includes little or no effect, the correct conclusion may be:
The evidence is inconclusive.
That is different from saying:
The campaign had zero effect.
Lack of sufficient evidence is not proof of no effect.
This distinction matters.
It prevents false confidence.
Integrated Campaigns Create a Serious Double-Counting Problem
The previous article in this series focused on connecting channels.
That same integration makes measurement harder.
A customer may see:
online video;
social media;
creator content;
retail media;
and an in-store placement.
Several platforms may then claim credit for the same purchase.
If channel-level attributed revenue is simply added together, the total can exceed actual brand sales.
That means campaign measurement should separate two questions.
Campaign-Level Question
Did the overall marketing intervention create incremental business impact?
Channel-Level Question
How much did each channel contribute to that impact?
The second question is usually harder.
In many cases, it may be more reliable to estimate campaign-level incrementality first and then use:
additional experiments;
MMM;
or hybrid methods
to understand channel contribution.
Otherwise the organisation risks producing mathematically precise double counting.
A Dashboard Is Not the Same as Measurement
A brand may have an impressive real-time dashboard showing:
impressions;
clicks;
video completion;
site traffic;
search;
sales;
ROAS;
and conversion.
That means the company has a lot of data.
It does not automatically mean it understands causal impact.
NIQ’s 2026 work on incrementality makes a similar point: the challenge is often not lack of information but the difficulty of combining:
media;
sales;
price;
promotion;
digital shelf;
distribution;
and other market variables
into a credible estimate of impact.
A dashboard can answer:
What happened?
To answer:
What would have happened without the campaign?
we need a measurement design.
Think in Terms of an Evidence Ladder
Not every campaign needs the most expensive methodology.
But teams should understand what kind of claim each method can support.
| Method | What it mainly tells us | Strength for causal claims |
|---|---|---|
| Impressions, clicks, views | Was media delivered and interacted with? | Very low |
| Campaign-period sales | What happened during the period? | Low |
| Before/after comparison | What changed versus the past? | Low |
| Attributed sales | Which purchases matched attribution rules? | Low to moderate |
| Modelled baseline | What may have happened without the campaign? | Depends on model |
| Matched-market analysis | Did treated markets outperform comparable markets? | Moderate to strong |
| Randomised experiment | Did the intervention cause a difference? | Very strong when well designed |
| Marketing Mix Modelling | How may multiple drivers contribute at broader scale? | Depends on assumptions and data |
| Experiment + model calibration | Are model estimates consistent with causal evidence? | Often stronger |
The exact hierarchy is context-dependent.
But the general principle is robust:
The strength of the conclusion should never exceed the strength of the evidence.
A More Complete Measurement Framework for Food Campaigns
Instead of relying on one KPI, it can be useful to evaluate performance across several layers.
Layer 1: Was the Media Delivered?
Reach.
Frequency.
Viewability.
Audience delivery.
This tells us whether the planned media execution happened.
Layer 2: Did Anything Change in the Mind?
Awareness.
Ad recall.
Association.
Favourability.
Consideration.
This is the communication effect.
Layer 3: Did Behaviour Change?
Search.
Product-page visits.
Store-locator use.
Trial.
Add-to-cart.
This is intermediate behaviour.
Layer 4: Did We Create Incremental Sales?
This requires a credible counterfactual.
Layer 5: What Kind of Sales Did We Create?
New buyers?
Existing buyers?
Higher purchase frequency?
Stockpiling?
Cannibalisation?
Halo?
Layer 6: Was the Growth Economically Valuable?
Incremental revenue.
Incremental margin.
Media cost.
Promotional cost.
Production cost.
Retail fees.
Layer 7: Did the Effect Persist?
Repeat purchase.
Post-promotion behaviour.
Retention of new buyers.
Potential longer-term brand effects.
This creates a much more useful interpretation than:
“The campaign was successful.”
Successful in what way?
For whom?
For how long?
At what economic cost?
Not Every Campaign Needs Every Layer
A small social activation designed to introduce one recipe does not necessarily justify a sophisticated causal-sales study.
A national product launch with millions in investment probably does.
IAB’s guidance makes a similar point: measurement design should reflect:
the business objective;
the investment level;
the decision being made;
and the level of causal confidence required.
Professional measurement does not mean:
use the most complicated method possible.
It means:
Use evidence strong enough for the importance of the decision.
A Measurement Report Should End in a Decision
Suppose the evidence shows:
incremental sales were positive,
but most came from existing loyal buyers.
What should change?
Or:
incremental sales were moderate,
but new buyers showed strong repeat.
Should the campaign scale?
Or:
brand lift was strong,
but immediate sales did not move.
Was immediate sales impact actually the campaign objective?
Or:
retail media shows excellent attributed ROAS,
but an experiment suggests weak incrementality.
Should budget move?
Measurement becomes valuable when it improves one of these decisions:
Continue?
Scale?
Reduce?
Change message?
Change audience?
Reallocate channels?
Fix distribution?
Reconsider price?
Run another experiment?
A 70-page report that changes no decision may contain excellent analysis.
It is not necessarily an effective management tool.
15 Questions to Ask Before Saying “The Campaign Drove Sales”
1- What was the campaign’s primary objective before it launched?
2- Was sales the main KPI, or was the campaign primarily designed to change a brand metric?
3- Is the sales number total, attributed or incremental?
4- What is our estimate of what would have happened without the campaign?
5- Do we have a credible test and control design?
6- Were price and promotions accounted for?
7- Were distribution and availability changes included?
8- Were seasonality, holidays and category trends considered?
9- Could competitor activity explain part of the result?
10- Are attribution windows and ROAS definitions consistent across channels?
11- How much incremental sales came from new buyers?
12- Did the campaign cannibalise other products in the portfolio?
13- What happened after promotions or paid media ended?
14- Did incremental margin justify the total cost of the campaign?
15- What is the uncertainty around the estimate, and what decision should the result change?
If several of these questions cannot be answered, the campaign may still have produced useful signals.
But the organisation should be cautious with the sentence:
“The campaign caused this sales increase.”
The Purpose of Measurement Is Not to Prove That Marketing Worked
This may be the most important principle in the entire discussion.
If campaign measurement exists mainly to produce a success story, the system will naturally search for:
the highest reach;
the strongest ROAS;
the largest sales increase;
the best engagement rate.
The real purpose of measurement is different:
Reduce uncertainty about what actually worked.
Sometimes the conclusion will be:
the campaign performed strongly.
Sometimes:
the media worked, but availability constrained sales.
Sometimes:
sales rose, but advertising was not the main cause.
Sometimes:
sales fell, but the campaign prevented a much larger decline.
And sometimes the most professional conclusion is:
“With the available evidence, we cannot confidently isolate the campaign effect.”
That is not a measurement failure.
It is the difference between analysis and storytelling.
The Better Question Is Not “How Much Did We Sell?”
Sales matters.
But the more complete question is:
How much of the sales was genuinely created by the campaign, what kind of sales was it, and did that value persist?
Answering that requires us to:
separate total sales from attributed sales;
separate attributed sales from incremental sales;
build a credible counterfactual;
account for price, promotion, distribution, seasonality and competition;
run experiments where practical;
use models with explicit assumptions;
measure brand and sales outcomes according to campaign objectives;
and evaluate the quality of the customers and revenue created.
In that system, success is no longer:
a green arrow on the final presentation slide.
Success is being able to say, with defensible evidence:
what changed;
by how much;
with what level of uncertainty;
why we believe the campaign contributed;
and most importantly:
what we will do differently because we learned it.
That is the point at which measurement stops being campaign reporting.
It becomes a learning system for growth.
Sources
- IAB & IAB Europe — Guidelines for Incremental Measurement in Commerce Media, 2025.
- Referenced for the principles of counterfactual measurement, bias control and selection among experimental, model-based and hybrid approaches.
- IAB / MRC — Retail Media Measurement Guidelines.
- Referenced for measurement transparency, attribution windows and standardisation of retail-media metrics.
- NielsenIQ — retail media and incrementality measurement guidance, 2026.
- Referenced for the distinction between attribution and causality and the importance of price, promotion, distribution and retail data when estimating incremental impact.
- Circana — research on Sales Lift and Incremental ROAS, 2025.
- Referenced for separating attributed sales from incremental sales and for buyer-level analysis such as new-customer contribution.
- Google Research — geo-experiment methodology.
- Referenced for randomised geographic experiments and the estimation of incremental return on advertising investment.
- Google Meridian — causal-inference and MMM documentation.
- Referenced for the role of model assumptions, confounders and experimental calibration in Marketing Mix Modelling.
- Brodersen et al. — Inferring Causal Impact Using Bayesian Structural Time-Series Models, Annals of Applied Statistics.
- Referenced for model-based estimation of counterfactual outcomes when direct experimentation is unavailable.
- Circana — HiPRO case study.
- Used as an example of combining brand-lift measurement with geographic holdout testing for incremental sales analysis; campaign outcomes are specific to that case.
- Kantar — brand lift and media-effectiveness research.
- Referenced for the distinction between immediate sales effects, brand outcomes and different measurement horizons.